Behavior Analysis of the Binary Hyperbolic Tangent (Btanh) Algorithm
Document Type
Conference Proceeding
Date of Original Version
1-1-2023
Abstract
This paper seeks to analyze the behavior and potential uses of the binary hyperbolic tangent (Btanh) algorithm [1], an algorithm designed to provide the activation function in a specific type of stochastic artificial neuron. Differences between Btanh outputs when used with and without an associated parallel counter are explored, and output formulas for each case are estimated. The question of whether the Btanh module could be decoupled from a parallel counter and be useful for stochastic computing applications outside of neural networks is discussed as well. It is determined that the output bitstreams from a parallel counter leads to unique behavior from the Btanh module, as opposed to when the Btanh inputs are generated independently, and this must be taken into account when decoupling the two modules.
Publication Title, e.g., Journal
Midwest Symposium on Circuits and Systems
Citation/Publisher Attribution
Shively, Seth, Eugene Chabot, John Dicecco, and Scott Koziol. "Behavior Analysis of the Binary Hyperbolic Tangent (Btanh) Algorithm." Midwest Symposium on Circuits and Systems (2023). doi: 10.1109/MWSCAS57524.2023.10406118.